Large Language Models in FinTech: A Systematic Review of Applications, Methodologies, Challenges, and Future Research

Abstract

Large language models (LLMs) have moved rapidly into financial technology, where they are used to analyze disclosures, answer questions over financial reports, support customers, and assist risk and compliance work. The research is spread across natural language processing, finance, and enterprise engineering, and it has not been brought together systematically. This article presents a systematic review of LLMs in FinTech, following PRISMA 2020 reporting guidance. It addresses four research questions covering applications, methodologies, challenges, and future research directions. Nine studies published between 2021 and 2024 were synthesized. They include financial benchmarks and datasets, domain-adapted and instruction-tuned financial models, an empirical study of LLMs processing corporate disclosures, surveys of foundation models and financial LLMs, and engineering studies on prompt design, guardrails, and workflow integration. The review finds that financial analysis and question answering dominate current research. Domain adaptation and benchmark development are the most mature methodologies. Numerical reasoning over tables and text, reliability, safety, and governed deployment are the main open challenges. The review proposes an integrative framework linking applications, methodologies, and controls, and sets out an agenda for future research focused on evaluation in realistic settings, customer-facing safety, and integration with risk governance.